Part II — Python for Quantitative Finance¶
This is not a general Python course. There are hundreds of those, and almost none of them prepare you for quantitative work. This part covers only what quantitative research and trading systems actually need: fast numerical computation, time-series manipulation at scale, disciplined code structure, asynchronous market data access, durable storage with point-in-time correctness, and the logging and configuration habits that make research reproducible. Anything a working quant does not routinely touch — web frameworks, GUI toolkits, generic scripting trivia — is deliberately absent.
The treatment is opinionated in the way professional desks are opinionated. Vectorization is not a performance nicety; it is the difference between a research iteration taking seconds or hours. Type hints and dataclasses are not bureaucracy; they are how research code survives contact with a second person, including your future self. Every lesson works with market data from the first example, so the idioms you learn are the ones you will use daily: aligning timestamps, resampling bars, handling NaNs deliberately, and storing data so that yesterday's query gives yesterday's answer.
By the end of this part you will have a working research stack — NumPy, Pandas/Polars, SQL, plotting, and a project structure with logging and configuration — that the statistics and strategy parts of the course build on directly. The mini-projects below stitch the individual lessons into an end-to-end pipeline.
Modules¶
| Module | Focus |
|---|---|
| NumPy and Vectorization | Arrays, broadcasting, vectorized computation, and numerical pitfalls |
| Pandas and Polars | Time-series DataFrames, resampling, rolling windows, joins, and lazy frames |
| Typing, Dataclasses, and Code Structure | Type hints, dataclasses, enums, and structuring research code with mypy |
| Async and Market Data APIs | async/await, REST vs streaming data, rate limits, and retries |
| SQL and Data Storage | Schema design for market data, SQLite to PostgreSQL, Parquet, point-in-time correctness |
| Plotting for Research | Equity curves, drawdowns, distributions, heatmaps, and IC-ready charts |
| Logging, Configuration, and Reproducibility | Structured logging, config management, and reproducible runs |
Mini-projects¶
- Market data downloader — download historical market data from a public API, with rate limiting and retry handling.
- Vectorized indicators — compute a set of standard indicators over a large universe using vectorized NumPy/Pandas, with no Python loops.
- OHLCV database — design a schema and store OHLCV data in SQL with appropriate keys and indexes.
- Query and resample — query historical data from the database and resample it to multiple bar frequencies without lookahead.